GCNG

GCNG infers extracellular gene–gene interactions from spatial transcriptomics and single-cell expression data using graph convolutional neural networks (GCNNs) to identify intercellular communication.

Key Features:

• Graph Convolutional Neural Network Approach: Extends graph convolutional neural networks (GCNNs) to process graph-encoded spatial information for gene interaction inference.
• Integration of Spatial and Expression Data: Encodes spatial information into a graph format and integrates it with single-cell expression data via supervised training to identify intra- and inter-cellular interactions.
• Handling Sparse and Noisy Data: Manages sparse and noisy expression vectors typical of spatial transcriptomics datasets.
• Multi-cell Type Profiling: Processes data from multiple cell types to enable inference in heterogeneous tissues.
• Neighborhood Definition: Defines cellular neighborhoods within the graph structure to reflect spatial proximity for extracellular interaction inference.
• Novel Interaction Proposals: Proposes novel pairs of extracellular interacting genes and has demonstrated ability to surpass previous methods in discovering such interactions.

Scientific Applications:

• Extracellular communication mapping: Infers intercellular gene–gene interactions to map cell–cell signaling networks from spatial transcriptomics data.
• Tissue development: Identifies extracellular interactions relevant to developmental processes in tissues.
• Immune responses: Reveals extracellular signaling relationships involved in immune cell communication.
• Cancer progression and tumor microenvironment: Detects intercellular interaction patterns that can inform studies of tumor microenvironment and cancer progression.

Methodology:

Data encoding of spatial information into graph structures; supervised training of graph convolutional neural networks to integrate spatial and single-cell expression data; inference of extracellular gene–gene interactions from the trained model; downstream analysis including functional assignment of inferred interactions.